Safety analysis of unsignalized intersections: a bivariate extreme value approach

ABSTRACT

Application of extreme value theory (EVT) to road safety analysis is gaining interest, thanks to its ability to produce quick and reliable safety evaluations without the use of crash data. Traditionally applied to single collision types and single extreme variables (i.e. surrogate measures of safety), EVT can be further exploited to simultaneously model multiple collision types, with the use of multiple extreme variables. In this paper two bivariate EVT approaches are applied for the safety evaluation of a three-leg unsignalized intersection, considering: (i) two conflict points and a single surrogate measure of safety; (ii) two surrogate measures of safety collected in a single conflict point. Each bivariate analysis was applied with two EVT methods: Component-wise Maxima (CM) and Excesses Over a Threshold (EOT). Bivariate models produced good results, especially with the EOT method, and were able to significantly improve the univariate benchmark results when the two estimation datasets were correlated.

KEYWORDS:

• Road safety
• extreme value theory
• unsignalized intersection
• collision prediction
• driving simulator
• crash probability
• accident prevention
• surrogate measures of safety

Acknowledgments

The authors thank Giulia Gaita and Stefano Bellan for their support during experiment execution and data collection.

Disclosure statement

The authors declare no conflict of interest.

Additional information

Funding

This work was supported by the University of Padova (project: ‘Advances in proactive and reactive approaches to road safety: development of innovative crash risk prediction and automatic incident detection systems’. Grant number: BIRD177377/17).